Jason Stowe, Cycle Computing | AWS Re:Invent 2013
summary
Cycle Computing provisions 156,000 cores across eight AWS regions using Amazon Spot to run 2.3 million compute hours in 18 hours, testing 205 candidate solar materials for efficiency. They combine OpsCode/Chef deployment with a custom Jupiter scheduler to distribute millions of parallel tasks and tolerate instance or region loss. The run costs about $30,000 versus a $68 million equivalent purchase and would take 12–18 months to provision on-premises.
AI-written summary · about
